botcorpus is not a chatbot or a search engine. It is a machine-callable knowledge layer: every fact carries a source, valid-from date and confidence — so an AI agent can rely on a verified value instead of guessing.
Tax rates, deadlines, wages, office holders — these sit past the training cutoff and change over time. An LLM will fill them in confidently, but without a source and often wrong. A single wrong rate or missed deadline costs an agent's task more than thousands of calls.
An AI agent or RAG pipeline needs a fact — a tax rate, a deadline, a municipality by postal code.
MCP or REST returns structured JSON with the value and the full trust envelope.
The agent acts on a verified value and can cite it — including the source and valid-from date.
Every fact, regardless of category, carries the same trust fields. With them an agent knows not only what, but also where from, how long it holds and how much to trust it.
// GET /facts/sk/contributions/employee-rate { "id": "sk/contributions/employee-rate", "value": 14.4, "unit": "%", "valid_from": "2026-01-01", "confidence": "high", "status": "ok", "source": "Podnikajte.sk — payroll contributions 2026", "source_url": "https://podnikajte.sk/...", "sources_count": 3 }
The detailed methodology — how we verify, date and version facts — is on the Data quality → page
Native call from an AI agent (Claude, n8n, your own runtime). The agent gets botcorpus as a tool and calls facts directly at runtime.
tool: botcorpus.lookup
Classic HTTP calls for backends and RAG pipelines. Filter by category, pull a single fact or look it up by key.
GET /facts/{id}
Download a whole category at once (JSON / CSV) and index it locally. For offline RAG and regular sync.
GET /datasets/{category}
Inventory of every corpus domain. Sorted by fact count: the most covered topics are at the top, and thinner areas show the gaps. Every fact has a primary source, valid-from date and confidence.
Free tier: 1,000 calls per month, no card. Connect via MCP and make your first verified call in a few minutes.